Schedule¶
| DAY 1 - Wednesday 29/05 | |
|
09:00 CEST
10:00 EEST |
Welcome and Introduction
Presenters: Jørn Dietze (LUST) and Christian Schou Oxvig (LUST and DeiC) |
|
09:15 CEST
10:15 EEST |
Introduction to LUMI
Presenter: Jørn Dietze (LUST) |
|
09:45 CEST
10:45 EEST |
Using the LUMI web interface
Presenters: Mats Sjöberg (CSC) and Lukas Prediger (CSC) |
|
10:05 CEST
11:05 EEST |
Hands-on: Run a simple PyTorch example notebook |
|
10:35 CEST
11:35 EEST |
Break (25 minutes) |
|
10:50 CEST
11:50 EEST |
Your first AI training job on LUMI
Presenters: Mats Sjöberg (CSC) and Lukas Prediger (CSC) |
|
11:20 CEST
12:20 EEST |
Hands-on: Run a simple single-GPU PyTorch AI training job |
|
12:05 CEST
13:05 EEST |
Lunch break (45 minutes) |
|
12:50 CEST
13:50 EEST |
Understanding GPU activity & checking jobs
Presenter: Samuel Añtao (AMD) |
|
13:10 CEST
14:10 EEST |
Hands-on: Checking GPU usage interactively using rocm-smi |
|
13:30 CEST
14:30 EEST |
Running containers on LUMI Presenter: Christian Schou Oxvig (LUST & DeiC) |
|
13:50 CEST
14:50 EEST |
Hands-on: Pull and run a container |
|
14:50 CEST
15:50 EEST |
Break (15 minutes) |
|
14:25 CEST
15:25 EEST |
Building containers from conda/pip environments Presenter: Christian Schou Oxvig (LUST & DeiC) |
|
14:45 CEST
15:45 EEST |
Hands-on: Creating a conda environment file and building a container using cotainr |
|
15:05 CEST
16:05 EEST |
Extending containers with virtual environments for faster testing Presenter: Gregor Decristoforo (LUST) |
|
15:25 CEST
16:25 EEST |
Getting started with your own project |
|
16:25 CEST
17:25 EEST |
End of the course day |
| DAY 2 - Thursday 30/05 | |
|
09:00 CEST
10:00 EEST |
Scaling AI training to multiple GPUs
Presenters: Mats Sjöberg (CSC) and Lukas Prediger (CSC) |
|
09:30 CEST
10:30 EEST |
Hands-on: Converting the PyTorch single GPU AI training job to use all GPUs in a single node via DDP |
|
10:00 CEST
11:00 EEST |
Hyper-parameter tuning using Ray on LUMI Presenter: Gregor Decristoforo (LUST) |
|
10:20 CEST
11:20 EEST |
Hands-on: Hyper-parameter tuning the PyTorch model using Ray |
|
10:40 CEST
11:40 EEST |
Break (15 minutes) |
|
10:55 CEST
11:55 EEST |
Extreme scale AI
Presenter: Samuel Añtao (AMD) |
|
11:25 CEST
12:25 EEST |
Demo/Hands-on: Using multiple nodes |
|
11:45 CEST
12:45 EEST |
Loading training data from Lustre and LUMI-O Presenter: Harvey Richardson (HPE) |
|
12:00 CEST
13:00 EEST |
Lunch break (60 minutes) |
|
13:00 CEST
14:00 EEST |
Coupling machine learning with HPC simulation Presenter: Harvey Richardson (HPE) |
|
13:30 CEST
14:30 EEST |
Advancing your own project |
|
16:00 CEST
17:00 EEST |
End of the course day |